Neural Scalable Symbolic Search for Complex Knowledge Graph Queries
A new framework called Neural Scalable Symbolic Search (NS3) addresses the challenge of answering complex existential first-order queries with multiple free variables (EFO_k) over incomplete knowledge graphs. Existing methods rely on marginal rankings over individual variables, which poorly approximate true joint rankings of answer tuples. NS3 extends neural symbolic search from EFO_1 queries to handle k free variables without enumerating the entire entity set E^k. It uses a budgeted approach that answers marginalized sub-queries to obtain candidate sets, merges multiple free variables into hypernodes, and approximates joint ranking efficiently. The framework is designed to scale as k grows, making it suitable for real-world knowledge graphs where exact enumeration is intractable. The paper is published on arXiv with ID 2605.25985.
Key facts
- NS3 is a budgeted framework for answering EFO_k queries over incomplete knowledge graphs.
- It approximates joint ranking of answer tuples without enumerating E^k.
- Existing methods use marginal rankings, which are poor proxies for joint rankings.
- NS3 merges multiple free variables into hypernodes to reduce complexity.
- The framework extends neural symbolic search from EFO_1 to EFO_k queries.
- The paper is available on arXiv under ID 2605.25985.
- Complex Query Answering (CQA) is a fundamental task in knowledge representation.
- NS3 answers marginalized sub-queries to obtain candidate sets.
Entities
Institutions
- arXiv